Repository navigation
Open
Description
Goal
Provide two genuine, independently executed consumer exercises before the separately authorized 0.6.0 launch. Neither internal CI nor the #90 published-vs-published onboarding demo can be counted as an external consumer; #90 counts as 0/2.
This issue is a test protocol and evidence ledger, not a claim that anyone has tested 0.6.0 yet. Refs #88 #86 #53.
Consumer A — new, independent FastAPI app
- Record an application/repository outside this project, with the independent runner or maintainer; avoid counting the fork's
examples/, CI fixtures, or two venvs from one app as separate consumers. - Install exact test artifact into a fresh environment, verify
fastapi_cacheresolves insidesite-packages, record source SHA and wheel/sdist SHA-256 pluspip check. - Exercise InMemory and real standalone Redis (on an isolated test DB/prefix): fresh endpoint MISS → HIT, TTL, clear(key/namespace), ETag/304 and lifecycle/close.
- Negatives: Authorization/Cookie requests must not leak, and an endpoint with tenant/user/locale/middleware/ContextVar-dependent results must either bypass or have a complete explicit identity-aware key; do not claim automatic isolation from
Varyalone. - Capture Python/FastAPI/Pydantic/redis-py/server version and redis standalone vs cluster/sentinel. Report incompatibilities and any installation wrinkles.
Consumer B — real legacy-to-fork migration
- Independent application previously using
fastapi-cache2==0.2.2, or a realistic separately maintained reproducer supplied and run by an external developer; don't count an internal repo-generated test as independent. - Record prior dependency lock, endpoint and namespace inventory (including private response/coder/custom key/multi-init cases), scope potential exposed legacy private cache entries without touching live production data.
- Run before-and-after in separate fresh environments (never co-install both distributions); prove default-key compatibility on ordinary public data and confirm intentionally diverging auth/request behavior.
- For a dedicated Redis test instance, demonstrate legacy writer → fork reader, selective cleanup of unsafe/private entries (unrelated keys survive), TTL/ETag, and rollback of package installation with private caching disabled/mitigated. No
FLUSHDBon shared Redis. - Capture artifact hashes, source SHA, Python/FastAPI/Pydantic, redis-py/server and redacted summary of actual breakages or code changes.
Privacy and evidence rules
- Never request credentials, PII, real tokens, raw Redis values from private namespaces or production logs.
- For each consumer, record independent project, named/consenting contact or anonymous identifier (private by default), date, exact release/candidate artifact, pass/fail, reproducible setup, mitigations and any unresolved high-severity concerns. Get explicit permission before attributing or quoting feedback publicly.
- Passing the internal migration/rollback workflow is necessary but not equivalent to independent external adoption. If only internally executed cases are available, report 0/2 rather than manufacturing evidence.
- Before 0.6.0 final sign-off, re-run any consumer result against the exact final candidate artifact/SHA, or clearly label older results as historical and non-certifying.
Launch gate
- Consumer A accepted with reproducible evidence.
- Consumer B accepted with reproducible evidence.
- All security/privacy failures triaged under security(cache): residual hidden tenant/identity context and header-cardinality safety after 0.5.0 #53; regression fixes completed or explicit NO-GO for unsafe library-controlled disclosure.
- 0.6.0 release acceptance references these two evidence records, not just a checkbox or unverified testimonial.
Do not solicit a publication date or user endorsement. No external outreach has been performed by creating this issue.
Metadata
Metadata
Assignees
Labels
No labels
Activity
docs(migration): runnable published before-and-after adoption demo (#90)